Heuristic Planning for Hybrid Systems

A High-Altitude Pseudo-Satellite (HAPS) is a fixed-wing, solar-powered Unmanned Aerial Vehicle (UAV) developed to become a flexible alternative to satellites with fixed-orbits for monitoring ground activities over long periods of time. However, given its lightweight build and weak electro-motors, the platform is rather sensitive to weather and cannot fly around hazardous weather zones swiftly. In this work, we formulate the problem of planning missions for multiple HAPS as a hybrid planning problem expressed in PDDL+. The formulation also considers the problem of modeling the platform dynamics, the time-varying environment, and the heterogeneous tasks that need to be carried out. Additionally, we propose a framework that combines a PDDL+ automated planner with an Adaptive Large Neighborhood Search (ALNS) approach, developed to couple the automated planner with a meta-heuristic that is specific for the problem. The task and motion planning are done in an intertwined way within the framework, preserving hence a common decision/search space. We validate our approach with a third-party realistic simulator for HAPS, as well as with a set of benchmark tests, showing that our integrated approach produces executable mission plans.

Section: Discussionmentioning

confidence: 99%

An AI-Based Planning Framework for HAPS in a Time-Varying Environment

Kiam

Scala

Javega

et al. 2020

“…We chose these planners because they take input in standard PDDL, and have been shown to outperform other planners, e.g. dReach (Piotrowski et al 2016); which are more difficult perform a fair comparison with due to the need to translate PDDL to drm.…”

Section: Discussionmentioning

confidence: 99%

“…The first is to discretize time according to a user-selected quantum . Planners such as UP-Murphi (Della Penna et al 2009), DiNo (Piotrowski et al 2016) and ENHSP (Scala et al 2016) take this approach. They are flexible in terms of the types of non-linear numeric functions supported, but depend heavily on an value selection that permits both scalability and solving of problems.…”

Section: Introductionmentioning

confidence: 99%

Mixed Discrete Continuous Non-Linear Planning through Piecewise Linear Approximation

Denenberg

Coles

2019

Reasoning with continuously changing numeric quantities is vital to applying planners in many real-world scenarios. Several planners capable of doing this have been developed recently. Scalability remains a challenge for such planners, especially those that reason with non-linear continuous change. In this paper, we present a novel approach to reasoning with non-linear domains. Bounding the problem using linear over and under-estimators, allows us to use scalable planners that handle linear change to find plans for non-linear domains. We compare the performance of our approach to existing planners on several domains and demonstrate that our planner can achieve state-of-the-art performance in non-linear planning.

“…Several PDDL extensions such as PDDL2.1 (Fox and Long 2003) and PDDL+ (Fox and Long 2006) support planning with numeric variables that evolve over time. Most numeric planners are limited to problems with linear or polynomial dynamics (Hoffmann 2003;Bryce et al 2015;Cashmore et al 2016); however, some planners can handle non-polynomial dynamics by discretizing time (Della Penna et al 2009;Piotrowski et al 2016). While it may be technically possible to analytically model, for example, collision constraints among 3D meshes using PDDL+, the resulting encoding would be enormous, far exceeding the capabilities of numeric planners.…”

Section: Related Workmentioning

confidence: 99%

PDDLStream: Integrating Symbolic Planners and Blackbox Samplers via Optimistic Adaptive Planning

Garrett

Lozano-Pérez

Kaelbling

2020

Many planning applications involve complex relationships defined on high-dimensional, continuous variables. For example, robotic manipulation requires planning with kinematic, collision, visibility, and motion constraints involving robot configurations, object poses, and robot trajectories. These constraints typically require specialized procedures to sample satisfying values. We extend PDDL to support a generic, declarative specification for these procedures that treats their implementation as black boxes. We provide domain-independent algorithms that reduce PDDLStream problems to a sequence of finite PDDL problems. We also introduce an algorithm that dynamically balances exploring new candidate plans and exploiting existing ones. This enables the algorithm to greedily search the space of parameter bindings to more quickly solve tightly-constrained problems as well as locally optimize to produce low-cost solutions. We evaluate our algorithms on three simulated robotic planning domains as well as several real-world robotic tasks.